you got that right. Point 1: Memristors are radically new technology still a long way from being reduced to practice. Point 2: The concept of memristors address memory. Period. The promised benefit is important. But memristors have NOTHING to do with synaptic connections - at least not in how we currently understand neuronal function. - For example, we are still grappling with the fact that a typical neuron will have thousands of other neurons connecting to it - firing at different times, with different neurotransmitters, with different intensity levels of firing. We are not even close to emulating a fraction of that capability. Side Point: symmetric versus asymmetric neural nets. As far as I know, we still have not successfully modeled asymmetric neural nets. If that is the case, it pushes the dramatic claim in the title even further into the future. Lots of sci fi novels to write just here...
Two further problems at that point become "nature vs. nurture" -- take the exact same "virgin" neural network and pre-load certain concepts into it.
a) can the concepts be overridden completely, or are they retained though distorted (good luck finding metrics to quantify this)
b) how does the 'auto didactic mode' ...you know, as it is assumed to work in humans ... have to do with socialization, as well as input such as TV, radio, and books.
Can one introduce a language and cognition of all nouns and verbs, without value judgments? (The temptation to cheat on such experiments, as in voter fraud in Dem strongholds, would be *enormous*.)